THREE LANGUAGES. EVERY NUANCE.

Listen closely.
See what changed.

Real utterances, heard again. Explore emotion and energy, listen to the audio used by the pipeline, and compare every word of its new transcript.

EnglishEspañolहिन्दी

THE REVIEW SETSTRATIFIED SAMPLE

—utterances

—detected emotions

3energy ranges

3target per cell

Original or cleaned audio, selected by the cleaner’s actual decision for each segment.

IN THIS VIEW—Loading review set

WORD DISAGREEMENT —word-weighted WER vs. original

EDIT COMPOSITION

—

AUDIO PROVENANCE—cleaned segments

Refine the collection

SEARCH TRANSCRIPTS

DETECTED EMOTIONENERGY RANGEAUDIO SOURCE

Sampling coverage

Clips per emotion × energy cell.
Target: 3 per language.

Filled cells Count / target

Disagreement distribution

Current curated selection.
Not population prevalence.

How to read this collection

WER is a comparison. The original transcript is the reference, not a verified human transcript. Higher disagreement can reflect a correction, a new error, or both.

WER = (substitutions + deletions + insertions) ÷ original reference words. Insertions can make WER exceed 100%. The overview is weighted by reference word count.

Word alignment ignores capitalization and punctuation while preserving language-specific letters and marks. Source annotations are excluded from the comparison; the verbatim source text is available in the word alignment panel when present.

Energy thresholds are determined from a source sample.

This is a balanced review sample, not an estimate of population emotion frequencies.

THE COLLECTION

Utterances worth a closer listen

Sort by

↔Read the original and new transcript side by side. Open Word alignment to inspect substitutions, deletions and insertions.

Loading the utterance collection…

Listen closely.See what changed.

Utterances worth a closer listen

Listen closely.
See what changed.